Food webs provide a framework for integrating population dynamics, community structure and ecosystem processes, but the causes of many food-web patterns remain controversial(1). An unresolved issue concerns the factors that limit food-chain length in natural ecosystems, although theory predicts that length could be limited by inefficient energy transfer between trophic levels(2) or by the instability of longer food chains(3). Experiments provide qualitative support for the instability of longer food chains(4), but, according to theory, this relationship may depend on the assumptions used to model food chains(5). Here we show that experimental manipulations of productivity determine the length of microbial food chains in laboratory microcosms. Food-chain length is also predicted to determine how productivity influences population densities within trophic levels. If trophic levels consist entirely of edible prey, and if consumers feed only on the next-lower trophic level(6), population densities should alternate between increasing and constant values as food-chain length increases (Fig. 1), We find that food-chain length determines population-level responses to productivity, which is consistent with predictions from models. These results indicate that the impacts of human alterations of productivity will depend crucially on the structure of the affected food chains.